使用MATLAB从图像中识别不同的硬币面值

Question

我正在尝试使用MATLAB识别图片中每个价值的火柴和硬币数量。这是原始图片，包含5个小银币、2个小金币、2个大银币和4个大金币。 输出结果如下： 以下是代码：

close all;
img = (imread('C:\Users\Torstein\Jottacloud\Skole\Visu\Prosjekt\sample_images\sample2.jpg'));
img_gray = rgb2gray(img);

% Filter image for easier edge detection
m = 12;
n = 12;
img_filter = imfilter(img_gray, fspecial('average', [m n]));
%figure, imshow(f), title('f')

% Edge detection
[~, threshold] = edge(img_filter, 'canny');
fudgeFactor = 1.5;
img_edge = edge(img_filter, 'canny', threshold * fudgeFactor);
figure, imshow(img_edge), title('edge detection')

% Dilate image to make the coin edges complete without holes
se_disk = strel('disk',4);
se_line1 = strel('line',3,100);
se_line2 = strel('line',3,100);
img_dilated = imdilate(img_edge, se_disk);
img_dilated = imdilate(img_dilated, [se_line1 se_line2]);
figure, imshow(img_dilated), title('dilate')

% Remove small objects (noise) and fill complete objects
img_clearborder = imclearborder(img_dilated, 4);
%figure, imshow(BWclear), title('cleared border image');
img_fill = imfill(img_clearborder, 'holes');
figure, imshow(img_fill), title('fill holes')

% Erode image to make a clear cut between objects
se_diamond = strel('diamond',2);
img_erode = imerode(img_fill,se_diamond);
for k=1:3
    img_erode = imerode(img_erode,se_diamond);
end
img_nosmall = bwareaopen(img_erode,300);
figure, imshow(img_nosmall), title('erode')

[B, L] = bwboundaries(img_nosmall);
figure, imshow(label2rgb(L, @jet, [.5 .5 .5])), title('boundaries')
hold on
for k = 1:length(B)
  boundary = B{k};
  plot(boundary(:,2), boundary(:,1), 'w', 'LineWidth', 2)
end

stats = regionprops(L,img(:,:,1),...
    'Area','Centroid','Orientation','EquivDiameter','MeanIntensity');
threshold = 0.80; % For differentiating coins from matches based on an objects circularity

coinCentroids = [];
coinIntensities = [];
matchCentroids = [];
matchAngles = [];
coinRatios = [];

for k = 1:length(B)
    boundary = B{k};
    delta_sq = diff(boundary).^2;
    perimeter = sum(sqrt(sum(delta_sq,2)));
    area = stats(k).Area;
    metric = 4*pi*area/perimeter^2;
    metric_string = sprintf('%2.2f',metric);
    angle_string = sprintf('%2.2f',stats(k).Orientation);
    centroid = stats(k).Centroid;
    if metric > threshold
        % Object is round, therefore a coin
        coinCentroids = [coinCentroids; centroid];
        coinIntensities = [coinIntensities; stats(k).MeanIntensity];
        coinRatios = [coinRatios; stats(k).EquivDiameter/area];
    else
        % Object is a match
        angle = stats(k).Orientation;
        matchCentroids = [matchCentroids; centroid];
        matchAngles = [matchAngles; angle];
    end

    plot(centroid(1),centroid(2),'ko');
%     text(boundary(1,2)-35,boundary(1,1)+13,angle_string,'Color','y',...
%       'FontSize',14,'FontWeight','bold');

end

如您所见，我已经确定了哪些物品是硬币，哪些是火柴。

但是，我很难确定硬币的价值。

例如，硬币的面积/直径给出以下结果。仅根据这些数据，我无法清晰地区分不同类型的硬币； 数字太接近了。

0.0041
0.0042
0.0043
0.0043
0.0044
0.0045
0.0048
0.0048
0.0053
0.0054
0.0055
0.0055
0.0056

我试着从每个硬币的起始图片中获取平均颜色强度，但这并没有帮助我区分银色硬币和金色硬币。

从红色通道得到的平均强度并不能告诉我们有6个金色硬币和6个银色硬币。

  105.0104
  105.4408
  107.9070
  112.4762
  116.3412
  127.3481
  132.1418
  137.9697
  149.6601
  159.2506
  167.6910
  181.1673
  215.0395

问题：我如何确定不同的硬币价值？

(在这里询问如何分离两个重叠的对象：使用MATLAB分离图像中的两个重叠圆形)

谢谢

- tsorn

---

我无法看到图片。错误403。 - Autonomous

尝试使用属性“'Image'”的regionprops函数，应该会给您一个很好的分离效果。 - Shai

@tos 另一个可能有帮助的数量是每个对象直径和面积之间的比率。 - Shai

你已经非常接近答案了，所以我不会提供答案。继续Shai说的话题，查看这篇文章：http://www.mathworks.com/matlabcentral/answers/85363#answer_94853 。有一个很好的公式可以计算物体的圆度。如果该值接近1，则更接近于圆形，而小于1则更不像圆形。使用公式中的参数与regionprops一起计算出圆度，然后使用类似于0.5的阈值提取出圆形对象。该帖子还提示要谨慎使用离心率，如果您决定使用它。 - rayryeng

使用硬币上的孔作为参数来确定价值是否可接受？我的意思是，假设您知道哪些硬币包含孔，因此您首先通过存在孔来对硬币进行分类，并进一步通过面积\直径进行细化 - 这可能足够了。要查找内部孔，可以使用Hough变换查找特定半径（我相信孔的半径约为5px）... - Dev-iL

Answer 1

首先，使用regionprops 'BoundingBox'，我使用imcrop和识别到的硬币的BoundingBox从起始图片中切出了硬币的图像。

然后，使用imfindcircles，我可以检测出银色硬币上的孔洞。最后，我使用硬币的面积来确定硬币的价值。

最终代码：

close all;
img = (imread('C:\Users\Torstein\Jottacloud\Skole\Visu\Prosjekt\sample_images\sample1.jpg'));
%figure, imshow(img);
img_gray = rgb2gray(img);

% img_hsv = rgb2hsv(img); 
% imgv = img_hsv(:,:,3);
% [Gx, Gy] = imgradientxy(imgv);
% [Gmag, Gdir] = imgradient(Gx, Gy);
% Gmag could be useful

% Filter image for easier edge detection
m = 12;
n = 12;
img_filter = imfilter(img_gray, fspecial('average', [m n]));
%figure, imshow(f), title('f')

% Edge detection
[~, threshold] = edge(img_filter, 'canny');
fudgeFactor = 1.5;
img_edge = edge(img_filter, 'canny', threshold * fudgeFactor);
%figure, imshow(img_edge), title('edge detection')

% Dilate image to make the coin edges complete without holes
se_disk = strel('disk',4);
se_line1 = strel('line',3,100);
se_line2 = strel('line',3,100);
img_dilated = imdilate(img_edge, se_disk);
img_dilated = imdilate(img_dilated, [se_line1 se_line2]);
%figure, imshow(img_dilated), title('dilate')

% Remove stuff touching the image border and fill complete objects
img_clearborder = imclearborder(img_dilated, 4);
%figure, imshow(BWclear), title('cleared border image');
img_fill = imfill(img_clearborder, 'holes');
%figure, imshow(img_fill), title('fill holes')

% Erode image to make a clear cut between objects
se_diamond = strel('diamond',2);
img_erode = imerode(img_fill,se_diamond);
for k=1:3
    img_erode = imerode(img_erode,se_diamond);
end
img_nosmall = bwareaopen(img_erode,300); % Remove small objects (noise)
%figure, imshow(img_nosmall), title('erode')

[B, L] = bwboundaries(img_nosmall);
%figure, imshow(label2rgb(L, @jet, [.5 .5 .5])), title('boundaries')
% hold on
% for k = 1:length(B)
%   boundary = B{k};
%   plot(boundary(:,2), boundary(:,1), 'w', 'LineWidth', 2)
% end

stats = regionprops(L,img(:,:,1),...
    'Area','Centroid','Orientation','EquivDiameter','Image','BoundingBox');
threshold = 0.80; % For differentiating coins from matches based on an objects circularity

coinCentroids = [];
coinTypes = []; % 0 for Silver, 1 for Gold
coinValues = []; % 1, 5, 10 eller 20 kroning
coinAreas = [];
silverCoinAreas = [];
goldCoinAreas = [];
matchCentroids = [];
matchAngles = [];
radiusRange = [8,40];

for k = 1:length(B)
    boundary = B{k};
    delta_sq = diff(boundary).^2;
    perimeter = sum(sqrt(sum(delta_sq,2)));
    area = stats(k).Area;
    metric = 4*pi*area/perimeter^2;
    metric_string = sprintf('%2.2f',metric);
    angle_string = sprintf('%2.2f',stats(k).Orientation);
    centroid = stats(k).Centroid;
    if metric > threshold
        % Object is round, therefore a coin
        coinValues = [coinValues; 0];
        coinAreas = [coinAreas; area];
        coinCentroids = [coinCentroids; centroid];
        bbox = stats(k).BoundingBox;
        im = imcrop(img,bbox);
        %figure, imshow(im);
        [centers,radii] = imfindcircles(im,radiusRange,'ObjectPolarity','bright');
        %viscircles(centers,radii);
        if length(centers) > 0
            % Coin has a hole, therefore either 1-kroning or 5-kroning
            coinTypes = [coinTypes; 0];
            silverCoinAreas = [silverCoinAreas; area];

        else
            % Coin does not have hole, therefore either 10-kroning or
            % 20-kroning
            coinTypes = [coinTypes; 1];
            goldCoinAreas = [goldCoinAreas; area];
        end

    else
        % Object is a match
        angle = stats(k).Orientation;
        matchCentroids = [matchCentroids; centroid];
        matchAngles = [matchAngles; angle];
    end

    %plot(centroid(1),centroid(2),'ko');
%     text(boundary(1,2)-35,boundary(1,1)+13,angle_string,'Color','y',...
%       'FontSize',14,'FontWeight','bold');

end

goldThreshold = 0.1;
silverThreshold = 0.1;
maxSilver = max(silverCoinAreas);
maxGold = max(goldCoinAreas);
for k=1:length(coinTypes)
    area = coinAreas(k);
    if coinTypes(k) == 0
        if  area >= maxSilver-maxSilver*silverThreshold
            % 5-kroning
            coinValues(k) = 5;
        else
            % 1-kroning
            coinValues(k) = 1;
        end
    else
        if area >= maxGold-maxGold*goldThreshold
            % 20-kroning
            coinValues(k) = 20;
        else
            % 10-kroning
            coinValues(k) = 10;
        end
    end
end

% OUTPUT:
coinCentroids
coinValues
matchCentroids
matchAngles

谢谢

- tsorn

---

干得好！这个线程是一个很好的提问示例，也展示了学习过程的结果。 - Shai